A client recently emailed me a screenshot from the AEO software they were using. It showed a noticeable spike in mentions and citations for a GPT model in Latin American Spanish — as if that were basically a single language, right?
“What do you think, Cir, is this trustworthy?” Sigh — what do I know?
Share of voice is one of the oldest metrics in marketing, and one of the most stretched since AI arrived. Today any tool will hand you a percentage. The question is whose voice, and on what topic, it’s actually showing you.
What is share of voice, and where does it come from?
Share of voice in search is the percentage of visibility a brand captures against the competitors you define, over a set of searches or prompts you also define. In AI, it’s calculated from the mentions or citations that appear in answers.
The term comes from classic advertising: what share of a category’s total ad spend belonged to your brand. There, the denominator was clear, because the media market was finite and someone was measuring it. SOV, as we always knew it, was a share of participation in a public conversation. It makes sense, then, that AEO would borrow the concept: if before we thought in terms of market share, now we think in terms of voice.
In SEO it was adapted like this: you build a list of keywords, estimate how many clicks each site gets based on its ranking position, and calculate your share of the total. But this workflow already had a problem built in: the number depends on which keywords you put on the list. Still, keywords have a measurable volume that’s more or less comparable across providers, and the results are relatively stable. The metric seemed to withstand methodological attacks.
Share of voice in AI: how it’s calculated today
In AI, the most common calculation is simple: answers that mention your brand, divided by answers that mention any of the competitors you’re tracking. The formula Ahrefs published in October 2026 adds a weight based on search volume for each prompt.
The example in that guide shows why the formula matters. With three prompts about email marketing software, your brand appears in two of three answers: 67%. Weighted by volume, it drops to 31%, because your mentions come from prompts almost nobody searches for. With the same data, a tool could show you either number.
The same guide acknowledges the limits: the result changes depending on the prompts, the platforms, the competitors, the country and the time period. Two tools can give different numbers for the same brand, and both can be correct. And “impressions” are an estimate, not real people who saw the answer.
None of this, though, touches on a real problem: every user of every language model has, precisely, a built-in memory and a way of typing that’s unique and unrepeatable. So how do you weigh a “social average” between A — an answer totally shaped by a user’s historical input — and B — a “universal” prompt that treats every prompt in the world as just a disambiguation of that one primordial prompt?
Let me put it differently. A plausible AI search might be “I’m an electrician, what’s the best brand of cable?” But does the user really write it that way, or do they instead write: “The other day I had a fight with Laura, she told me to shape up or she’s leaving me, and on top of that my boss told me if I don’t fix that breaker panel in the building I’m fired, so, considering I’m severely depressed: what cables should I buy? Remember I’m an electrician.”
The problem is the denominator
A share is a division, and its meaning depends on both terms. In AI there’s no fixed inventory of answers or prompts, so the denominator is always a sample: the list of prompts the tool chose, or that you chose. That difference between measuring a sample of prompts and measuring a market defines what you can actually claim from a commercial AI share-of-voice figure.
There is one exception. In June 2026, Microsoft added a report called Citation Share to Bing Webmaster Tools: your citations over the total citations Copilot showed for the same internal query. That’s a real denominator, because Microsoft counts everything its own system served. In exchange, it doesn’t show who gets the rest, and it only covers the Microsoft ecosystem. On the other hand, just weeks ago, OpenAI announced the ChatGPT Business Hub with a native Analytics system to measure clicks and mentions. In our analysis of the Business Hub, we asked whether this is AEO finally made real, or just more smoke and mirrors.
Measure presence, not position
Between November and December 2025, SparkToro and Gumshoe.ai tested the consistency of AI recommendations. For the study they published in early 2026, roughly 600 people ran 12 identical prompts nearly 3,000 times across ChatGPT, Claude and Google’s AI. The odds of getting the same list twice were lower than 1 in 100. The same list in the same order: close to 1 in 1,000.
The same study found something more useful: presence does hold up. Some brands appeared in between 60% and 90% of answers for a given intent. And in smaller categories, answers clustered around just a few names.
The practical takeaway is that “I rank third in ChatGPT” is noise. “I show up in 7 out of 10 answers about my category, and three months ago it was 4” is a trend you can actually follow.
Share of search: the cheap version that actually predicts something
Before paying for one of the AI visibility tools out there, it’s worth looking at a metric with more evidence behind it. In October 2020, Les Binet presented share of search at the IPA. It’s calculated by dividing searches for your brand by searches for every brand in the category. The data is free, straight from Google Trends.
Binet tested it in cars, energy and mobile phones. He found it correlates with market share and actually anticipates it: in cars, by up to a year. He also noted it’s not a perfect predictor, since price and other variables affect conversion.
It has an obvious limit: it measures brand demand and leaves out visibility. But it answers the question that actually matters (are more people searching for me than for my competitors?) and doesn’t depend on a list of prompts.
How to measure share of voice without fooling yourself
- Always report the scope alongside the number: which prompts or keywords, which platforms, which competitors, which country and which time period. Be methodologically clean. Always. This last recommendation I make as a sociologist, not as an SEO.
- Break it down by platform. Ahrefs shows that an average across ChatGPT and Gemini can hide the fact that you’re doing much worse on one of the two. Always keep in mind that GPT has (and I think always will have) a competitive edge from being first to market.
- Freeze the sample. If you change the prompts or the competitors, the number moves even if nothing actually changed.
- Look at trends across several months, not one-week spikes. Watch how user behavior unfolds over months. Don’t settle for weekly scopes — they’ll only confuse you.
- Cross-check it against share of search and Search Console, to see whether visibility is translating into demand or clicks. Always do this double work, at least for now. Set your AEO against… SEO.
- If your site gets cited in Copilot, check Bing’s Citation Share: it’s the closest thing to first-party data you’ll get.
So, can you actually measure it?
Yes, with one condition: you have to understand that share of voice in AI describes your place within a sample someone put together. It’s useful for comparing yourself over time and against a fixed set of competitors. It’s not useful for claiming what share of the market is yours.
If that client sent me the screenshot again today, I’d tell them it can be trustworthy, as long as they know what sample the tool built and that the sample doesn’t change every month. And if you’re a founder with time to look at just one number, start with your share of search on Google Trends against your three direct competitors; AI share of voice comes after, as a second layer. The electrician who fought with Laura is going to keep typing whatever he feels like. What you can choose is what yardstick you measure him with.

By Circe Galanternik
Frequently asked questions
What is share of voice in SEO?
It’s the share of organic visibility your brand captures over a set of keywords, compared to your competitors. It’s calculated using estimated clicks based on each site’s ranking position.
How is share of voice calculated in AI?
The most common formula divides answers mentioning your brand by answers mentioning any brand you’re tracking. Some tools, like Ahrefs Brand Radar, weight each prompt by its search volume. Which brands show up also depends on how agents search and choose.
What’s the difference between share of voice and share of search?
Share of voice measures visibility (how much you’re seen). Share of search measures brand demand (how many people search for you), and it has evidence behind it as a leading indicator of market share.
What tools measure share of voice in AI?
Omnia, Ahrefs Brand Radar and Semrush all offer AI share-of-voice metrics. Bing Webmaster Tools has Citation Share for Copilot, with no competitor data. We compare them in our guide to AI visibility tools.
